Automotive System

Spatial AI is crucial for autonomous vehicle safety but high computational power consumption reduces driving range. Our research focuses on energy-efficient methods using HW-SW co-design and sensor-friendly AI SoCs to enhance autonomous driving efficiency and safety.

DNN Accelerator

Hand gesture recognition (HGR) using 3D-CNNs and super resolution (SR) with deep neural networks both demand significant computational power, posing challenges for real-time processing on mobile devices. Dedicated ASIC processors are needed to enable efficient HGR and SR on battery-limited devices.

Processing-in-Memory

Convolutional Neural Networks (CNNs) excel in image and video processing but are limited by high computational and memory demands. Computing-in-Memory (CIM) architecture offers a solution by processing data within on-chip memory, significantly enhancing throughput and energy efficiency for ultra-low-power IoT devices.

Neuromorphic

The human brain consumes only 20 mW with 1 billion neurons in computation. Spiking Neural Networks (SNNs) mimic the behavior of biological neural networks to reduce power consumption of Artificial Neural Networks (ANNs). Neuromorphic processors accelerates SNNs for ultra low power hardware such as always-on-sensors, surveilance monitoring, bio-sensor back-end.